Allow processes to idle before exiting

When system load exceeds the ability of os_process_soft_limit to keep
up with demand we enter a fork-use-kill (FUK) cycle. The constant
spawning and destruction os these processes thrashes system resources
and causes general instability.

This patch changes the behavior from killing each process as its
returned to letting it idle for a configurable amount of time (default
five minutes) which allows it to be reused by other clients. This way we
can avoid adding unnecessary load when demand for couchjs processes
exceeds os_process_soft_limit.

As a happy benefit this should also allow os_process_soft_limit to be
set much lower since the number of processes will now more closely
follow actual demand (instead of provisioning for the worst case
scenario).

Conflicts:

	apps/couch/src/couch_os_process.erl
	apps/couch/src/couch_proc_manager.erl

Conflicts:
	apps/couch/src/couch_os_process.erl

2 files changed, 42 insertions, 26 deletions

diff --git a/apps/couch/src/couch_os_process.erl b/apps/couch/src/couch_os_process.erl
index 2a6d92a7..0c6f284f 100644
--- a/apps/couch/src/couch_os_process.erl
+++ b/apps/couch/src/couch_os_process.erl
@@ -27,7 +27,8 @@
      port,
      writer,
      reader,
-     timeout=5000
+     timeout=5000,
+     idle
     }).
 
 start_link(Command) ->
@@ -104,12 +105,15 @@ readjson(#os_proc{} = OsProc) ->
 
 % gen_server API
 init([Command, Options, PortOptions]) ->
+    V = couch_config:get("query_server_config", "os_process_idle_limit", "300"),
+    IdleLimit = list_to_integer(V) * 1000,
     Spawnkiller = filename:join([code:priv_dir(couch), "couchspawnkillable.sh"]),
     BaseProc = #os_proc{
         command=Command,
         port=open_port({spawn, Spawnkiller ++ " " ++ Command}, PortOptions),
         writer=fun writejson/2,
-        reader=fun readjson/1
+        reader=fun readjson/1,
+        idle=IdleLimit
     },
     KillCmd = readline(BaseProc),
     Pid = self(),
@@ -131,32 +135,32 @@ init([Command, Options, PortOptions]) ->
             Proc#os_proc{timeout=TimeOut}
         end
     end, BaseProc, Options),
-    {ok, OsProc}.
+    {ok, OsProc, IdleLimit}.
 
 terminate(_Reason, #os_proc{port=Port}) ->
     catch port_close(Port),
     ok.
 
-handle_call({set_timeout, TimeOut}, _From, OsProc) ->
-    {reply, ok, OsProc#os_proc{timeout=TimeOut}};
-handle_call({prompt, Data}, _From, OsProc) ->
+handle_call({set_timeout, TimeOut}, _From, #os_proc{idle=Idle}=OsProc) ->
+    {reply, ok, OsProc#os_proc{timeout=TimeOut}, Idle};
+handle_call({prompt, Data}, _From, #os_proc{idle=Idle}=OsProc) ->
     #os_proc{writer=Writer, reader=Reader} = OsProc,
     try
         Writer(OsProc, Data),
-        {reply, {ok, Reader(OsProc)}, OsProc}
+        {reply, {ok, Reader(OsProc)}, OsProc, Idle}
     catch
         throw:{error, OsError} ->
-            {reply, OsError, OsProc};
+            {reply, OsError, OsProc, Idle};
         throw:{fatal, OsError} ->
             {stop, normal, OsError, OsProc};
         throw:OtherError ->
             {stop, normal, OtherError, OsProc}
     end.
 
-handle_cast({send, Data}, #os_proc{writer=Writer}=OsProc) ->
+handle_cast({send, Data}, #os_proc{writer=Writer, idle=Idle}=OsProc) ->
     try
         Writer(OsProc, Data),
-        {noreply, OsProc}
+        {noreply, OsProc, Idle}
     catch
         throw:OsError ->
             ?LOG_ERROR("Failed sending data: ~p -> ~p", [Data, OsError]),
@@ -164,16 +168,23 @@ handle_cast({send, Data}, #os_proc{writer=Writer}=OsProc) ->
     end;
 handle_cast(stop, OsProc) ->
     {stop, normal, OsProc};
-handle_cast(Msg, OsProc) ->
+handle_cast(Msg, #os_proc{idle=Idle}=OsProc) ->
     ?LOG_DEBUG("OS Proc: Unknown cast: ~p", [Msg]),
-    {noreply, OsProc}.
+    {noreply, OsProc, Idle}.
 
+handle_info(timeout, #os_proc{idle=Idle}=OsProc) ->
+    gen_server:cast(couch_proc_manager, {os_proc_idle, self()}),
+    erlang:garbage_collect(),
+    {noreply, OsProc, Idle};
 handle_info({Port, {exit_status, 0}}, #os_proc{port=Port}=OsProc) ->
     ?LOG_INFO("OS Process terminated normally", []),
     {stop, normal, OsProc};
 handle_info({Port, {exit_status, Status}}, #os_proc{port=Port}=OsProc) ->
     ?LOG_ERROR("OS Process died with status: ~p", [Status]),
-    {stop, {exit_status, Status}, OsProc}.
+    {stop, {exit_status, Status}, OsProc};
+handle_info(Msg, #os_proc{idle=Idle}=OsProc) ->
+    ?LOG_DEBUG("OS Proc: Unknown info: ~p", [Msg]),
+    {noreply, OsProc, Idle}.
 
 code_change(_OldVsn, State, _Extra) ->
     {ok, State}.
diff --git a/apps/couch/src/couch_proc_manager.erl b/apps/couch/src/couch_proc_manager.erl
index 0f35422b..568608bc 100644
--- a/apps/couch/src/couch_proc_manager.erl
+++ b/apps/couch/src/couch_proc_manager.erl
@@ -67,13 +67,29 @@ handle_call({ret_proc, #proc{client=Ref, pid=Pid} = Proc}, _From, State) ->
     % #proc{} from our own table, so the alternative is to do a lookup in the
     % table before the insert.  Don't know which approach is cheaper.
     case is_process_alive(Pid) of true ->
-        maybe_reuse_proc(State#state.tab, Proc);
+        ets:insert(State#state.tab, Proc);
     false -> ok end,
     {reply, true, State};
 
 handle_call(_Call, _From, State) ->
     {reply, ignored, State}.
 
+handle_cast({os_proc_idle, Pid}, #state{tab=Tab}=State) ->
+    Limit = couch_config:get("query_server_config", "os_process_soft_limit", "100"),
+    case ets:info(Tab, size) > list_to_integer(Limit) of
+        true ->
+            ets:delete(Tab, Pid),
+            case is_process_alive(Pid) of
+                true ->
+                    unlink(Pid),
+                    gen_server:cast(Pid, stop);
+                _ ->
+                    ok
+            end;
+        _ ->
+            ok
+    end,
+    {noreply, State};
 handle_cast(_Msg, State) ->
     {noreply, State}.
 
@@ -95,7 +111,7 @@ handle_info({'DOWN', Ref, _, _, _Reason}, State) ->
         ok;
     [#proc{pid = Pid} = Proc] ->
         case is_process_alive(Pid) of true ->
-            maybe_reuse_proc(State#state.tab, Proc);
+            ets:insert(State#state.tab, Proc);
         false -> ok end
     end,
     {noreply, State};
@@ -110,17 +126,6 @@ terminate(_Reason, #state{tab=Tab}) ->
 code_change(_OldVsn, State, _Extra) ->
     {ok, State}.
 
-maybe_reuse_proc(Tab, #proc{pid = Pid} = Proc) ->
-    Limit = couch_config:get("query_server_config", "os_process_soft_limit", "100"),
-    case ets:info(Tab, size) > list_to_integer(Limit) of
-    true ->
-        ets:delete(Tab, Pid),
-        unlink(Pid),
-        exit(Pid, kill);
-    false ->
-        garbage_collect(Pid),
-        ets:insert(Tab, Proc#proc{client=nil})
-    end.
 
 get_procs(Tab, Lang) when is_binary(Lang) ->
     get_procs(Tab, binary_to_list(Lang));